EP0695989B1 - Inversion circuit for galois field elements - Google Patents

Description

The present invention relates to circuits allowing perform calculations on elements of a Galois body, and in particular a circuit making it possible to calculate the inverses of these elements.

A Galois body is a finite set of numbers binaries used, for example, to correct errors in data transmission thanks to Reed-Solomon coding and decoding.

The set of binary numbers of n bits forms a Galois field of 2 ⁿ = N + 1 elements, on which we define an internal addition and multiplication, i.e. such as the sum or the product of two body numbers is also a body number. The addition of two numbers consists in performing an exclusive OR bit by bit between these two numbers. From which it follows that, if x is any element of the body, we have x + x = 2x = 0.

Multiplication is a classic multiplication of two binary numbers of n bits as long as it does not generate a retained, i.e. as long as the result does not contain bits to 1 of weight greater than n-1. As soon as a holdback is generated, this is combined by exclusive OR with predetermined bits of bits of weight 0 to n-1, defined by a polynomial called polynomial Galois body generator.

Any non-zero element of the Galois body is a power of another non-zero and non-unitary element of this body. In a Galois body of N + 1 elements, these powers are defined modulo-N, that is to say that x ⁱ = x ^{i + N} , where x is a non-zero and non-unitary element of the Galois body and i a positive or negative integer. The elements of a Galois field of N + 1 elements are noted 0, α ⁰ , α ¹ , ... α ^N-1 . The elements α ⁰ to α ^N-1 are the numbers 2 ⁰ , 2 ¹ ... 2 ^N-1 constituting the base of the binary numbers of n bits.

To calculate correction coefficients in a Reed-Solomon decoder, it is necessary to calculate ratios y / x, where y and x are numbers calculated by the decoder and which can have any values. For this, we generally multiply y by the inverse of x.

To calculate an inverse, we can use a reverse table stored in a ROM memory. However, the use of a ROM memory does not lend itself well to integration among other processing circuits with techniques current integrated circuit design. With these techniques, the ROM memory must be placed outside an area where are integrated the other elements of the processing circuit. This results in a significant loss of surface despite a ROM memory occupying a relatively small area.

Another solution is to obtain the inverse of wired way, using logic gates. However, the number of connections between the logic gates to perform the function reversal is so important that the corresponding metallizations occupy an area equivalent to the lost area driven by use of ROM memory, despite the fact that the wired inverter can be integrated in the same area as the processing circuits.

Other solutions are described in the documents WO-A-89 01660 and GB-A-2 155 219.

An object of the present invention is to provide a inverter of elements of a Galois body occupying a surface particularly weak.

This object is reached by a circuit according to the claim 1.

According to an embodiment of the present invention, the elevator at the power t comprises n groups of doors of which the i-th (i = 0, 1 ... n-1), provides or not the i-th element not zero of the Galois field according to state 1 or 0 of the i-th bit of the number to be raised to the power t; and n-1 adders whose j-th (j = 1, 2 ... n-1) receives the output of the j-th group of doors
and the output of the (j-1) th adder, the first adder receiving the outputs of the first two groups of gates.

la figure 1 représente un circuit d'inversion ;

la figure 2 représente un circuit élévateur à la puissance d'une puissance de deux utilisé dans le circuit d'inversion de la figure 1 ;

la figure 3 représente un mode de réalisation de certains des éléments du circuit de la figure 1 ; et

la figure 4 représente un mode de réalisation selon la présente invention de certains des éléments du circuit de la figure 1.

These objects, characteristics and advantages as well as others of the present invention will be explained in more detail in the following description of particular embodiments given without limitation by means of the attached figures, among which:

Figure 1 shows an inversion circuit;

2 shows a booster circuit to the power of a power of two used in the inversion circuit of Figure 1;

Figure 3 shows an embodiment of some of the elements of the circuit of Figure 1; and

FIG. 4 represents an embodiment according to the present invention of some of the elements of the circuit of FIG. 1.

As described in particular WO-A-89 01660, the inverse x ^-1 of a number x of n bits is expressed in the form: x -1 = x t / x t + 1 , with t = 2 ^{n / 2} .

The number x ^t is particularly easy to calculate, as we will see later, thanks to the fact that it is a number raised to a power of a power of 2 (t = 2 ^{n / 2} ).

The number x ^{t + 1} is a root (t-1) th of the unit because (x t + 1 ) t-1 = x t 2 -1 = x 2n-1 = x NOT = 1. Consequently, x ^{t + 1} , whatever the value of x, takes only one of the t-1 values α t + 1 , α 2 (t + 1) ... α (t-1) (t + 1) . These t-1 values are noted below respectively β, β ² , ... β ^t-1 .

Thus, we are led to calculate the inverse of the number x ^{t + 1} which has only t-1 = 2 ^{n / 2} -1 possible values β, β ² , ... β ^t-1 instead of the 2 ⁿ -1 possible values of the inverse of any number. For example, if n = 8, the number x ^{t + 1} has 15 possible values instead of 255. Thus, an inverter of the number x ^{t + 1} occupies significantly less area than an inverter of any number, than it is carried out using a table in ROM memory, or using logic circuits. Furthermore, it suffices to provide this inverter only with bits of the number x ^{t + 1} which make it possible to distinguish the t-1 possible values from the number x ^{t + 1} .

FIG. 1 represents an inverter circuit obtained directly using the decomposition (1) above. The number x of n bits is supplied to an elevator at the power t 10. A multiplier 12 receives the output of the elevator 10 on a first input and the number x on a second input. An inverter 14 receives the output, x ^{t + 1} , of the multiplier 12 and provides the corresponding inverse, x ^{- (t + 1)} , to a first input of a multiplier 16. As previously indicated, the inverter 14 should only supply t-1 values. The multiplier 16 receives on a second input the output of the elevator 10 and provides the inverse x ^-1 sought.

The booster circuit 10 at the power t, t being a power of 2, is particularly simple to carry out for reasons explained below.

The number x is expressed by x = x 0 · α 0 + x 1 · α 1 + x 2 · α 2 + ... x n-1 · α n-1 , where x ₀ , x ₁ ... x _n-1 are the values of the bits of increasing weight of the number x.

By raising the number x to the power t, t being a power of 2, we raise the second member to the power t, which provides the sum of terms x _i · α ^it and additional terms which repeat an even number of time. Because the sum over the Galois field is a bit-by-bit exclusive OR, all of these additional terms cancel each other out. Thus, we have: x t = x 0 · α 0 + x 1 · α t + x 2 · α 2t + ... x n-1 · α (N-1) t .

FIG. 2 represents a booster circuit at the power t directly established from this equation. Each bit x _i (i = 0, 1 ... n-1) is associated with a group of AND gates 18 which receive the first bit x _i on the first inputs and the respective bits of the number α ^it on second inputs. Thus, each number α ^it is transmitted to the output of the corresponding group 18 if x _i = 1. A first adder 20 receives on a first input the output of group 18 associated with bit x ₀ and on a second input the output of group 18 associated with bit x ₁ . Additional adders 20 are associated respectively with the remaining groups 18 and each receives on a first input the output of the associated group 18 and on a second input the output of the previous adder 20. The last adder 20 provides the number x ^t .

Of course, in each group of AND gates 18 associated with a bit x _i , AND gates are provided for the only non-zero bits of the number α ^it . Likewise, the adders 20 (groups of exclusive OR gates) can also be simplified by taking into account the fact that some of their input lines are always in a constant state.

FIG. 3 represents an embodiment of a circuit intended to advantageously replace the inverter 14 and the multiplier 16 of FIG. 1. This circuit comprises n / 2 multipliers 22 by a constant. The multiplication constants are respectively the numbers β ²ⁱ , where i varies from 0 to n / 2-1.

Each multiplier 22 is associated with a multiplexer 24 which receives on a first channel the output of the multiplier and on a second channel the output of the preceding multiplexer, which is also supplied at the input of the multiplier. Using this arrangement, by properly selecting the multiplexers 24, a multiplication is obtained by any value β ⁱ , where i varies from 1 to t-1. The first multiplier 22 and first multiplexer 24 receive the number x ^t delivered by the elevator 10. The last multiplexer 24 provides the inverse x ^-1 sought. A decoder 26, receiving the number x ^{t + 1} supplied by the multiplier 12, controls the multiplexers 24 so that the multipliers 22 are placed in series, the product of the constants being equal to the inverse of x ^{t + 1} .

The circuit of Figure 3 is particularly simple because it uses multipliers by a constant; providing n / 2 with associated multiplexers remains simpler than providing a full multiplier 16. In addition, the decoder 26 has only n / 2 outputs instead of the n outputs required in the case of the figure 1. As in the case of FIG. 1, to control the decoder 26, it is only supplied with bits of the number x ^{t + 1} which make it possible to distinguish the t-1 possible values from the number x ^{t + 1} .

In certain applications, the structure of FIG. 3 might not be fast enough because the number x ^t must sometimes cross a large number of multipliers 22 which each introduce a delay.

FIG. 4 represents a particularly rapid embodiment according to the present invention. The number x ^t is supplied in parallel to t / 2 multipliers 40 by a constant. The constants are respectively β ₁ to β _½t-1 . The constants β ₁ to β _½t-1 are ½t-1 roots (t-1) th of unity distinct from each other and from unity, and such that the remaining ½t-1 roots are 1 + β ₁ to 1+ β _½t-1 . Indeed, on a Galois field, if a number r is p-th root of unity, the number r + 1 is also (p being any integer less than N + 1).

A multiplexer 42 receives the outputs of the multipliers 40 and selects one of them as a function of a control signal supplied by a decoder 44. The output of the multiplexer 42 is supplied to a first input of an adder 46 receiving on a second input the output of a group of AND gates 48. A first input of the gates 48 receives an output of the decoder 44, which takes a state dependent on the parity of the inverse of the number x ^{t + 1} .

The second inputs of gates 48 respectively receive the bits of the number x ^t . Thus, the number x ^t is summed or not at the output of the multiplexer 42 according to the values of the number x ^{t + 1} . With this configuration, we multiply x ^t by β _i or 1 + β _i (i = 1, 2 ... ½t-1), that is to say by one or the other of two roots (t -1) distinct themes of the unit. The decoder 44, as a function of the n-1 most significant bits of the number x ^{t + 1} selects the appropriate multiplier 40 so that the number β _i or 1 + β _i by which the number x ^t is finally multiplied is the inverse of the number x ^{t + 1} . In addition, the decoder receives only bits of the number x ^{t + 1} which make it possible to distinguish the ½t-1 roots β ₁ to β _½t-1 .

Of course, among the roots (t-1) th of unity, there is unity. If the number x ^{t + 1} is equal to 1, the number x ^t is also equal to 1, then, for example, the multiplexer 42 is selected to supply the number 0 to the adder 46 which then supplies the value 1 / ( x ^t ) through door group 48.

Claims (2)

1. A circuit for inverting a number (x) of n bits of a finite field of 2ⁿ = N+1 elements, comprising:

   a circuit (10) for raising to the power t = 2^n/2, receiving the number to invert (x);

   a first complete multiplier (12) receiving the number to invert and the output of the circuit for raising to the power t; and

   a circuit (14, 16) for providing the product of the output of the circuit for raising to the power t and of the inverse of the output of the first complete multiplier;

   characterized in that said circuit for providing the product comprises:

   1/2t-1 multipliers (40) for multiplying by a constant, whose constants are (t-1)th roots of the unit, distinct one from the other and from the unit, each receiving the output of the circuit (10) for raising to the power t;

   a multiplexer (42) controlled by a decoder (44) for selecting the output of one of the multipliers by a constant as a function of the output of the first complete multiplier (12), and

   an adder receiving the output of the multiplexer and, depending upon the output of the first complete multiplier, value 0 or the number to invert (x) raised to the power t.
2. The circuit of claim 1, characterized in that the circuit (10) for raising to the power t comprises:

   n groups of gates (18), the i-th group (i=0, 1... n-1) providing or not the i-th non-zero element of the finite field depending upon the state 1 or 0 of the i-th bit of the number to be raised to the power t; and

   n-1 adders (20), the j-th adder (j=1, 2... n-1) receiving the output of the j-th group of gates and the output of the (j-1)th adder, the first adder receiving the outputs of the first two groups of gates.

Images